Final Algorithm 1

Loading...

來自 University of California San Diego 的課程

Advanced Algorithms and Complexity

212 個評分

University of California San Diego

Advanced Algorithms and Complexity

212 個評分

課程 5（共 6 門，Specialization 数据结构与算法）

You've learned the basic algorithms now and are ready to step into the area of more complex problems and algorithms to solve them. Advanced algorithms build upon basic ones and use new ideas. We will start with networks flows which are used in more typical applications such as optimal matchings, finding disjoint paths and flight scheduling as well as more surprising ones like image segmentation in computer vision. We then proceed to linear programming with applications in optimizing budget allocation, portfolio optimization, finding the cheapest diet satisfying all requirements and many others. Next we discuss inherently hard problems for which no exact good solutions are known (and not likely to be found) and how to solve them in practice. We finish with a soft introduction to streaming algorithms that are heavily used in Big Data processing. Such algorithms are usually designed to be able to process huge datasets without being able even to store a dataset.

從本節課中

Streaming Algorithms (Optional)

In most previous lectures we were interested in designing algorithms with fast (e.g. small polynomial) runtime, and assumed that the algorithm has random access to its input, which is loaded into memory. In many modern applications in big data analysis, however, the input is so large that it cannot be stored in memory. Instead, the input is presented as a stream of updates, which the algorithm scans while maintaining a small summary of the stream seen so far. This is precisely the setting of the streaming model of computation, which we study in this lecture. The streaming model is well-suited for designing and reasoning about small space algorithms. It has received a lot of attention in the literature, and several powerful algorithmic primitives for computing basic stream statistics in this model have been designed, several of them impacting the practice of big data analysis. In this lecture we will see one such algorithm (CountSketch), a small space algorithm for finding the top k most frequent items in a data stream.

Introduction5:23

Heavy Hitters Problem7:33

Reduction 14:35

Reduction 26:30

Basic Estimate 18:04

Basic Estimate 27:14

Final Algorithm 15:27

Final Algorithm 212:57

與講師見面

Alexander S. Kulikov
Visiting Professor
Department of Computer Science and Engineering
Michael Levin
Lecturer
Computer Science
Daniel M Kane
Assistant Professor
Department of Computer Science and Engineering / Department of Mathematics
Neil Rhodes
Adjunct Faculty
Computer Science and Engineering

In the rest of the lecture, we'll put together the ideas that we developed for

our basic version of the approximate point query data structure,

to get a full float approximate point query.

And finally, the count sketch algorithm that gets good results,

good approximations to frequency counts with high probability and small space.

I will start by providing the intuition for the improvements that

we have to make to the original version of approximate point theory.

Then I will define the algorithm itself and finally give some intuition for

the analysis of the parameters.

Okay, we will recall that our basic version of ApproxPointQuery suffered from

two issues.

Well first, the precision of our estimates depended on the variance

of a certain estimator, and that variance sometimes was bad.

For example,

when several dominant coefficients were present in the same string.

To fix this deficiency, what we do for

our final algorithm is run our basic estimator that we already designed

on subsampled or hashed version of the stream of data items that we get us in.

The reason why this works is that this operation as we will show later

reduces variance of our estimates.

Now the second problem with our basic version of

ApproxPointQuery was that the success probability was fairly low.

It would only succeed with high constant probability.

On the other hand, our final algorithm has to provide good estimates for

all data items that appearing in streams simultaneously.

So to fix this deficiency and make the algorithm run with high probability,

we introduce the idea of repeating our estimates independently many,

many times and taking medians of answers that they report to boost our confidence.

Okay, so the main idea of this algorithm, and the first idea that makes this work,

is the idea of hashing the input stream into a number of buckets.

So again, here's our data items, we associate them with integers between

one and some large number, for our examples, we use 1 through 10.

What we do is we choose a hash function that maps the data items

to a number of buckets.

So in this particular case, on the slide you see a random mapping of data items 1

through 10 into 8 hash buckets, b is number of hash buckets, it's equal to 8.

The mapping is shown as the arrows, so for

example, element number 1 is mapped into bucket 3 and

element number 1 also happens to collide with element number 6 in bucket 3.

Element 10 on the other hand is residing alone in bucket 2,

it doesn't collide with any other element.

Good, now this idea of hashing let's us define the notion of hashed streams.

In particular, for every item i, we will say that the hashed stream of item i

contains all the other elements that collide with i under hash function h.

An example, the subsampled or hashed stream of item 1,

under this particular hash function shown on the slide consists of 2 items,

item 1 itself and item 6, because they collided in bucket 3.

The subsampled or hashed stream of item 5 consists of two items,

5 and 7, because 5 and 7 collided in bucket 5.

Now it is important to note that we are hashing the universe of

data items into a number of buckets, and not positions in the stream.

Let me give a specific example.

So this is the stream that we started with at the beginning of this lecture.

So it contains occurrences of items from 1 to 10, and

here is the frequency histogram.

Under the hash function from the previous slide, the subsample stream of item 1,

which is the heavy hitter, one of the dominant items,

most frequent item on this stream, is shown at the bottom of the slide.

It consists of all the occurrences on 1 in the stream and all the occurrences of 6.

The frequency histogram for the stream is shown at the bottom of the slide.

The crucial observation is that this frequency histogram looks exactly

like the good case for our basic estimator from the previous part of the lecture.

Indeed, if we were to estimate the frequency of item 1 from this particular

stream then the arrow would only be contributed by the item number 6.

Which means that we would actually get a good estimate,

because item number 1 dominates the subsampled or hash stream.

Okay, but item number 1 was a heavy hitter, it

was the most frequent item on the string, but see what happens for other items.

So for example if you look at the subsampled hashed stream of item number 5,

all that consists of two items, 5 and 7.

Just all occurrences of these items and 5 occurs twice and

7 occurs once, the frequency histogram again is shown at the bottom of the slide.

So note that if we were to estimate item number 5 from this particular stream,

we would actually get a very good estimate.

Because this estimate would not be contaminated by occurrences of any of

the heavy hitters in the original stream.

And this is the basic idea for

why our new version of the algorithm will actually work well.

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera 致力於普及全世界最好的教育，它與全球一流大學和機構合作提供在線課程。

© 2018 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取